Differential prediction performance between Caribbean- and Mainland-subgroups using state-of-the-art polygenic risk scores for coronary heart disease: Findings from the Hispanic Community Health Study/Study of Latinos (HCHS/SOL)

Background Coronary heart disease (CHD) is a leading cause of death for Hispanic/Latino populations in the United States. We evaluated polygenic risk scores (PRS) with incident myocardial infarction (MI) in a Hispanic/Latino study sample. Methods We leveraged data from the Hispanic Community Health Study/Study of Latinos (HCHS/SOL) to assess four CHD-PRS from the PGS catalog, derived using multiple methods (LDpred, AnnoPred, stacked clumping and thresholding, and LDPred2). We evaluated associations between each standardized PRS and time to adjudicated incident MI, adjusted for age, sex, first 5 principal components, and weighted for survey design. Concordance statistics (c-index) compared predictive accuracy of each PRS with, and in addition to, traditional risk factors (TRF) for CHD (obesity, hypercholesterolemia, hypertension, diabetes, and smoking). Analyses were stratified by self-reported Caribbean- (Puerto Rican, Dominican or Cuban) and Mainland- (those of Mexican, Central American, or South American) heritage subgroups. Results After 11 years follow-up, for 9055 participants (mean age (SD) 47.6(13.1), 62.2% female), the incidence of MI was 1.0% (n = 95). Each PRS was more strongly associated with MI among Mainland participants. LDPred2 + TRF performed best among the Mainland subgroup; HR=2.69, 95% CI [1.71, 4.20], c-index = 0.897, 95% CI [0.848, 0.946]; a modest increase over TRF alone, c-index = 0.880, 95% CI [0.827, 0.933]. AnnoPred + TRF performed best among the Caribbean sample; c-index = 0.721, 95% CI [0.647, 0.795]; however, was not significantly associated with rate of MI (HR=1.14, 95% CI [0.82, 1.60]). Conclusion PRS performance for CHD is lacking for Hispanics/Latinos of Caribbean origin who have substantial proportions of African genetic ancestry, risking increased health disparities. AnnoPred, using functional annotations, outperformed other PRS in the Caribbean subgroup, suggesting a potential strategy for PRS construction in diverse populations. These results underscore the need to optimize cumulative genetic risk prediction of CHD in diverse Hispanic/Latino populations.